We went from a 1.8% false positive rate to under 0.4% in two weeks. Our ops team stopped drowning in manual review queues. The contributing_signals[] response made it possible to explain every decision to the business.
Behavioral transaction scoring
Catch synthetic identities and account takeovers in 47ms.
Txnworks scores every transaction against 140+ behavioral signals — device fingerprint, velocity clustering, behavioral drift, identity consistency — before the payment clears. Rules-based systems miss 40% of what signal clusters catch.
LIVE FEED
~47ms median
Shopify #4821
$84.50
US
43ms
ALLOW
WooCommerce #9103
$12.00
CN
51ms
BLOCK
Stripe #7744
$329.00
US
38ms
ALLOW
Adyen #2255
$1.00
NG
46ms
REVIEW
Braintree #6631
$215.75
CA
41ms
ALLOW
Plaid #8840
$5.00
RU
49ms
BLOCK
Marqeta #3318
$76.00
GB
44ms
ALLOW
Galileo #5522
$2.50
US
37ms
REVIEW
fraud_rules.conf
IF velocity_24h > 5 THEN BLOCK
MISS
IF country NOT IN allowlist THEN REVIEW
MISS
IF amount > 500 AND new_device THEN BLOCK
MISS
IF email_domain IN blocklist THEN BLOCK
MISS
IF card_age_days < 30 THEN REVIEW
MISS
Rules miss what patterns see.
-
Synthetic identity fraud assembles real data fragments — no single rule catches the composite.
-
Account takeover looks identical to the real user until behavioral drift surfaces in signal clusters.
-
Rules debt compounds: every new attack vector needs a new rule, each one generating new false positives.
How Txnworks scores a transaction.
01
Ingest
POST /v1/score with transaction fields — REST or streaming webhook.
02
Signal compute
140+ signals across device fingerprint, velocity patterns, behavioral drift, and network reputation — computed in parallel.
03
Decision
Risk score 0-1000 returned in under 50ms. ALLOW / REVIEW / BLOCK with confidence band.
Four fraud patterns. One scoring call.
Synthetic Identity
Composites of real and fabricated PII that pass identity verification but fail behavioral consistency.
Account Takeover
Credential stuffing and session hijack flagged by device divergence and behavioral drift signals.
Card Testing
Low-value probes from stolen card lists — detected by velocity clustering and merchant pattern analysis.
Bust-Out Fraud
Gradual credit line exploitation with clean history until sudden high-value default — caught by timeline drift signals.
Works with your payment stack.
Stripe
Adyen
Braintree
Plaid
Marqeta
Galileo
REST / Webhook
REST API · Webhooks · SDKs: Node.js, Python, Go
What early users say.
The SDK was live in our staging environment the same afternoon. Score response included contributing signals with weights — our team could read exactly why each transaction scored the way it did. No black box, no vendor call to decode it.
Ready to stop chasing fraud rules?
Score your first transaction in under 15 minutes. No sales call required.
Request API Access